Anthropic mathematician claims to have proved the Riemann hypothesis with Claude
Levent Alpöge published his proof in Lean, and MIT’s Andrew Sutherland says he compiled it in Lean himself. Experts still need to check how the claim is stated in Lean and to understand what the proof contains.
Levent Alpöge is a mathematician at the AI company Anthropic. In a four-page text posted on his website on Friday 20 November, he wrote: ‘I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis.’ Claude is Anthropic’s large language model. Alpöge said the method could also solve the generalised Riemann hypothesis. Anthropic has not commented, The Verge and Nature reported.
The Riemann hypothesis, posed in 1859, concerns the distribution of prime numbers. It states that the non-trivial zeros of the Riemann zeta function all lie on the same vertical line. The hypothesis is one of the Clay Mathematics Institute’s Millennium Prize Problems, each of which carries a prize of 1 million dollars for its solution.
Alpöge published the proof in Lean. When a proof is written in Lean, a computer checks every step of the reasoning. MIT mathematician Andrew Sutherland told Scientific American that he had compiled the result in Lean himself. Other mathematicians have confirmed this, TechCrunch reported. The computer, however, checks only the statement written in the Lean file. Experts still need to verify whether the hypothesis has been stated correctly in Lean and whether the proof holds, Scientific American reported.
The proof is still hard for people to understand. Alpöge says Claude’s proof runs to more than 250,000 pages, and the Lean file contains millions of lines, SiliconANGLE reported. Andrew Granville, a number theorist at the University of Montreal, told Nature that the text Alpöge released was garbled. Sutherland said understanding the details of the proof still requires a great deal of work.
Anthropic went public the day before the publication. Granville links the timing of the publication to the fall in the company’s share price on 19 November, Nature reported. In his text, Alpöge rejects such a cynical reading.
On 10 August, Anthropic wrote that Claude had raised the lower bound on the fraction of the zeta function’s zeros that satisfy the hypothesis. The bound rose from 41.6 per cent to 67.2 per cent. At the time, the company did not expect Claude’s techniques to lead to a proof of the hypothesis.
The Clay Institute awards the prize only once a solution has been published, scrutinised for at least two years and generally accepted by the mathematical community, The Verge reported. Meanwhile, Stanford professor Ravi Vakil published a proposed proof of the generalised hypothesis by postdoctoral researcher Alessandro Masullo. Vakil said it was produced entirely without AI. The proposal has not been written in Lean, so it is too early to judge its validity, TechCrunch reported.